CN107704953A - The short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimate forests - Google Patents
The short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimate forests Download PDFInfo
- Publication number
- CN107704953A CN107704953A CN201710850654.7A CN201710850654A CN107704953A CN 107704953 A CN107704953 A CN 107704953A CN 201710850654 A CN201710850654 A CN 201710850654A CN 107704953 A CN107704953 A CN 107704953A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- mtd
- quantile
- mtr
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 69
- 230000009466 transformation Effects 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 28
- 238000003066 decision tree Methods 0.000 claims description 22
- 230000001419 dependent effect Effects 0.000 claims description 16
- 230000008569 process Effects 0.000 claims description 15
- 238000001228 spectrum Methods 0.000 claims description 13
- 238000004422 calculation algorithm Methods 0.000 claims description 11
- 230000008901 benefit Effects 0.000 claims description 9
- 238000005315 distribution function Methods 0.000 claims description 9
- 238000007637 random forest analysis Methods 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000005516 engineering process Methods 0.000 claims description 7
- 230000001186 cumulative effect Effects 0.000 claims description 6
- 230000011218 segmentation Effects 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 5
- 238000000205 computational method Methods 0.000 claims description 4
- 238000010801 machine learning Methods 0.000 claims description 4
- 101100391181 Dictyostelium discoideum forH gene Proteins 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 3
- 238000013524 data verification Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 230000000717 retained effect Effects 0.000 claims description 3
- 230000005856 abnormality Effects 0.000 claims description 2
- 238000010276 construction Methods 0.000 claims description 2
- 238000000354 decomposition reaction Methods 0.000 description 12
- 230000003044 adaptive effect Effects 0.000 description 6
- 230000005611 electricity Effects 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 230000013016 learning Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000012706 support-vector machine Methods 0.000 description 3
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 2
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 2
- 230000000875 corresponding effect Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000013277 forecasting method Methods 0.000 description 2
- 210000002816 gill Anatomy 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000000611 regression analysis Methods 0.000 description 2
- 238000012502 risk assessment Methods 0.000 description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 1
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 238000012300 Sequence Analysis Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000002592 echocardiography Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Separation Using Semi-Permeable Membranes (AREA)
Abstract
The present invention discloses a kind of short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimates forest, comprises the following steps:1) original wind power sequence is decomposed into by series of features inequality empirical mode using experience wavelet transformation (empirical wavelet transform, EWT);2) empirical mode is reconfigured according to frequency range, forms high frequency, intermediate frequency and low frequency component;3) input variable is chosen using Pearson correlation coefficients to each component;4) quantile estimate forest forecast model is established to each component, obtains different quantile regression forecasting results;5) each component prediction result is superimposed, obtains wind power prediction value;6) prediction of wind power probability density is obtained using Density Estimator.The method provided by the present invention effectively increases wind power prediction precision, obtains the prediction of any time wind power probability density, can preferably solve the problems, such as power system wind power prediction.
Description
Technical field
The present invention relates to a kind of power-system short-term wind power probability density Forecasting Methodology, to power system wind power
It is predicted, belongs to technical field of power systems.
Background technology
Wind-power electricity generation is installed the rising year by year of ratio in power network, effectively alleviates energy shortage, environmental pollution general layout, but
Its intermittent and uncertain safety and stability and economical operation that drastically influence power network again.Short-term wind-electricity power prediction is as certainly
Dynamic Generation Control and the important decision foundation for arranging power scheduling, can effectively improve Operation of Electric Systems reliability.Therefore, need
New technology and new method are studied, to improve wind power prediction precision, meets engineer applied demand.
At present, domestic and foreign scholars have carried out numerous studies, main having time sequence analysis, people to short-term wind-electricity power prediction
The models such as artificial neural networks, SVMs, Method Using Relevance Vector Machine.In addition, further to reduce wind power prediction error, it is related
Scholar proposes combination forecasting.It facts have proved:Built-up pattern can have complementary advantages relative to Individual forecast method, improve
While precision of prediction, model robustness is enhanced.Combined prediction is different by mechanism strategy, is broadly divided into two classes:1) using not
Forecast model with principle is predicted respectively, and prediction result then is optimized into combination by certain way;2) signal is used
Treatment technology carries out resolution process to original wind power sequence, forecast model is established to different decomposition amount, finally to each component
Prediction result is combined.Using wavelet transformation progress, data prediction is faced with wavelet basis selection, Decomposition order is difficult to
The problem of determining.Wind power sequence is automatically decomposed into by a series of intrinsic mode letters using adaptive empirical mode decomposition
, when then establishing forecast model to mode function, modal overlap phenomenon be present, influence precision of prediction in number.Integrated empirical modal point
Solution method effectively alleviates modal overlap problem existing for empirical mode decomposition, improves precision of prediction, but its amount of calculation it is larger,
Operation efficiency is low.
Easily occur the shortcomings of low modal overlap, computational efficiency, shortage theoretical foundation for empirical mode decomposition method,
Gilles propose NEW ADAPTIVE signal processing method-experience wavelet transformation (empirical wavelet transform,
EWT).This method combines the adaptive characteristic and Wavelet Analysis Theory framework of empirical mode decomposition method, by signal frequency
The adaptivenon-uniform sampling of spectrum, the amplitude modulationfrequency modulation of Fourier spectrum is extracted in the suitable orthogonal wavelet filter of each spectrum architecture
(amplitude modulated-frequencymodulated, AM-FM) composition, and then converted using Hilbert to difference
AM-FM mode handled, finally obtain significant instantaneous frequency and instantaneous amplitude.This method amount of calculation is small, and has
Stronger robustness.Therefore, EWT is introduced into short-term wind-electricity power prediction modeling by the present invention, and original wind power sequence is entered
Row resolution process.
In general short-term wind power forecast method only provides deterministic point prediction as a result, it is difficult to which wind energy is fully described
Uncertainty and changing rule, so as to be unfavorable for policymaker in Electric Power Network Planning operation, risk analysis, reliability assessment etc.
Make scientific and effective decision-making.Therefore, related scholar proposes short-term wind-electricity power probability forecasting method, as quantile estimate,
Interval prediction, density prediction etc..Probabilistic forecasting can preferably describe the possible fluctuation range of following wind power, uncertainty
And the risk faced.Nicolai combinations quantile estimate is theoretical, is proposed on the basis of random forest (random forest, RF)
Quantile estimate forest (quantile regression forests, QRF) model, different quantiles can be provided and returned
Prediction result.As a kind of nonparametric ensemble machine learning method, QRF has that arithmetic speed is fast, model performance is influenceed by parameter concurrently
Small, relatively strong the advantages that holding making an uproar property.The present invention establishes QRF wind power prediction models, is obtaining different quantile prediction output bars
Under part, the prediction of wind power probability density is further obtained using Density Estimator.
In summary, the present invention combines EWT and QRF advantages, establishes EWT-QRF short-term wind-electricity powers probability density prediction mould
Type.First, original wind power sequence is decomposed into by a series of different empirical mode of frequencies using EWT, to each Empirical Mode
Formula establishes QRF forecast models respectively, obtains different quantile regression forecasting results, and each empirical mode prediction result is superimposed, obtained
To final wind power prediction value.Finally, the prediction of wind power probability density is provided using Density Estimator method.
The content of the invention
Goal of the invention:The present invention is for problem present in existing Load Prediction In Power Systems technology, such as in general wind-powered electricity generation
Power forecasting method can only export deterministic point prediction as a result, it is difficult to reflect randomness, the uncertainty of wind power completely
Easily there is the problems such as low modal overlap, computational efficiency, shortage theoretical foundation shortcoming in feature, empirical mode decomposition method, there is provided one
The short-term wind-electricity power probability density Forecasting Methodology of kind EWT quantile estimate forests.The present invention uses EWT by original wind-powered electricity generation first
Power sequence is decomposed into the empirical mode of series of features inequality, and choosing input variable collection to each empirical mode merges foundation point
Digit returns forest forecast model, obtains the prediction result under the conditions of any quantile.Predicted by being superimposed different empirical modes
As a result, final short-term wind-electricity power predicted value is obtained.Finally, the condition distribution to predicted value is obtained using Density Estimator method
Obtain any time wind power probability density prediction result.
Technical scheme:A kind of short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimates forest, including it is following
Step:
(1) master data needed for power system short-term wind power prediction is obtained, and initial data is pre-processed,
Carry out abnormality value removing or amendment;
(2) pretreated wind power sequence is decomposed into the empirical mode of series of features inequality using EWT;
(3) empirical mode is reconfigured according to frequency range, forms high fdrequency component, intermediate frequency component and low frequency point
Amount;
(4) input variable set is chosen using Pearson correlation coefficients to each component;
(5) quantile estimate forest model is established to each component, obtains different quantile regression forecasting results;
(6) each component prediction result is superimposed, obtains final wind power prediction value;
(7) prediction of wind power probability density is obtained using Density Estimator, and by surveying wind power data verification
The validity of the inventive method.
Further, original wind power sequence is decomposed into a series of frequecy characteristic inequalities by step (2) using EWT technologies
Empirical mode, the specific calculating process of EWT methods is:
2.1 determine the frequency range of bandpass filter;Adaptivenon-uniform sampling, definition are carried out to the Fourier spectrums of signal first
Fourier supports is [0, π] and assume to be divided into N number of continuous part, make Λn=[wn-1,wn] represent each segmentation side
Boundary;Wherein:N=1,2, L, N, w0=0, wN=π, wnIn being chosen between the two neighboring maximum point of signal Fourier spectrums
Point, it is obvious
2.2 with each wnCentered on, definition width is Tn=2 τnTransitional region;
2.3 in segmentation section ΛnOn, it is each Λ to define experience small echonOn bandpass filter, and according to Meyer small echos
Building method constructs experience small echo, and the experience wavelet function generally constructed is
Experience scaling function is
In formula:τn=γ wn;When, it ensure thatFor tight frame;Generally, letter
Number β (x) is defined as β (x)=x4(35-84x+70x2-20x3);
2.4 so as to which primary signal can be reconfigured as
In formula:* it is convolution algorithm;For Coefficients of Approximation;For x (t) experience wavelet transformation;
Empirical mode xk(t) it is defined as
Further, step (3) is reconfigured empirical mode according to frequency range, forms high fdrequency component, intermediate frequency
Component and low frequency component, the frequency range demarcation method are specially:Empirical mode in [0, π] frequency range is divided into three
Part:Frequency being more than 1 for low frequency component, frequency and being more than 2 as high frequency division less than 2 for intermediate frequency component, frequency less than 1
Amount.
Further, step (4) chooses input variable set, the Pearson to each component using Pearson correlation coefficients
Coefficient correlation circular is:
Preset time sequence, Pearson correlation coefficients have weighed xtWith xt-τBetween dependency relation, can effectively determine defeated
Enter variables collection;Wherein, τ is lag order;Pearson correlation coefficients r spans are [- 1,1], and 1 represents perfect positive correlation ,-
1 represents perfect negative correlation, and 0 represents non-correlation between variable;When | r | during > 0.3, that is, think exist between variable compared with strong correlation pass
System;Coefficient correlation calculation formula is between moment wind power to be predicted and preceding τ moment power
In formula:xt,i、xt-τ,iRespectively variable xtAnd xt-τI-th of sample value;Respectively variable xtAnd xt-τ's
Average value;N is sample size.
Further, step (5) establishes quantile estimate forest model to each component, obtains different quantile regression forecastings
As a result;Quantile estimate forest model combines the advantage of quantile estimate and Random Forest model, can provide difference point position
Point regression forecasting result;As a kind of nonparametric ensemble machine learning method, QRF has that arithmetic speed is fast, model performance is joined concurrently
Number influences small, relatively strong the advantages that holding making an uproar property;The specific calculating process of quantile estimate is:
5.1 under specified criteria X=x, and conditional distribution function is Y≤y cumulative probability, i.e.,
F (y | X=x)=P (Y≤y | X=x)
α quantiles Qα(x) it is that Y is more than or equal to Q under the conditions of given X=xα(x) cumulative probability is exactly α, i.e.,
Qα(x)=inf { y:F (y | X=x) >=α }
5.2 in general linear conditions quantile estimates are expressed as
QY(α | X=x)=β0(α)+β1(α)x1+β2(α)x2+L+βk(α)xk≡X′β(α)
In formula:QY(α | X=x) is dependent variable Y in independent variable X=[x1,x2,L,xk] under the α condition quantile, point
Site α ∈ (0,1), β (α) is regression coefficient vector, and it changes with quantile α change;
5.3 condition quantiles solve parameter vector β (α) estimate by minimizing loss function, define loss function
For
5.4 can be converted into following optimization problem so as to quantile estimate
Give some quantile α, pass through parameter vector estimate corresponding to solution, you can describe independent variable now to because
The influence of variable;Then when α is in the continuous value in (0,1) feasible section, you can obtain Y condition distribution.
Further, step (5) establishes quantile estimate forest model to each component, obtains different quantile regression forecastings
As a result;The specific calculating process of quantile estimate forest is:
6.1 random forests are counted as an adaptability nearest neighbour classification and regression process, to each X=x, can obtain
Original one weight set w of n observed valuei(x), i=1,2, L, n;Random forest essence is to utilize all dependent variable observations
Estimation of the weighted sum as dependent variable Y conditional means E (Y | X=x);In addition, QRF decision trees are produced with standard RF algorithms,
Condition distribution is obtained by the dependent variable Weighted estimation observed, wherein the weight of each observation is equal to RF algorithm weights;
6.2 thus, QRF defines E (1{Y≤y}| X=x) be estimated as observation 1{Y≤y}Weighted average, i.e.,
6.3 k decision tree T (θ of generationt), t=1,2, L, k;To each leaf node of every decision tree, the leaf node is investigated
All observations;
6.4 given X=x, travel through all decision trees;Calculate the weight w of every decision tree observationi(x,θt), i ∈ 1,2,
L,n};By to decision tree weight wi(x,θt), t=1,2, L, k are averaged to obtain the weight of each observation i ∈ { 1,2, L, n }
wi(x);
6.5 couples of all y ∈ R, the weight drawn using step 6.4, you can calculate the estimation of distribution function;To every decision-making
Each node of tree, RF, which is returned, only to be retained the average of observation and have ignored other information, and QRF retains all observations in node
Value, and condition distribution is calculated on this basis.
Further, the prediction of wind power probability density is obtained using Density Estimator in step (7), and by surveying wind
The validity of electrical power data verification the inventive method;The specific calculating process of Density Estimator is:
Density Estimator is to estimate its density by the stochastic variable from same unknown distribution function of one group of observation
The nonparametric computational methods of function;If X1,X2,L,XnIt is taken from unitary continuous population sample, the total body density at the x of arbitrfary point
Function f (x) Density Estimator is defined as
In formula:K (x) is kernel function, the gaussian kernel function form used herein forH is bandwidth
Coefficient, span are 1.8~2.0.
Beneficial effect:The power-system short-term wind power probability density Forecasting Methodology of the present invention utilizes EWT by original wind
Electrical power sequence is decomposed into the empirical mode of series of features inequality, and input variable is chosen to each empirical mode and is established and divides position
Number returns forest forecast model, obtains the regression forecasting result under the conditions of any quantile, further obtains wind power probability
Density prediction.Relative in general Forecasting Methodology, the inventive method can provide the prediction of any time wind power probability density
As a result, scientific and effective determine is made in Electric Power Network Planning operation, risk analysis, reliability assessment etc. so as to be beneficial to policymaker
Plan.
Brief description of the drawings
Fig. 1 is split for Fourier spectrum;
Fig. 2 is original wind power sequence and EWT decomposition results;
Fig. 3 is empirical mode spectrum distribution;
Fig. 4 is empirical mode reconstruction result;
Fig. 5 is original wind power sequence and EMD decomposition results;
Fig. 6 is that EMD decomposes mode function reconstruct component;
Fig. 7 is 80% confidential interval probabilistic forecasting result;
Fig. 8 is 1:00 moment probability density is predicted;
Fig. 9 is 5:00 moment probability density is predicted;
Figure 10 is 9:00 moment probability density is predicted;
Figure 11 is 13:00 moment probability density is predicted;
Figure 12 is 17:00 moment probability density is predicted;
Figure 13 is 21:00 moment probability density is predicted.
Embodiment
With reference to specific embodiment, the present invention is furture elucidated, it should be understood that these embodiments are merely to illustrate the present invention
Rather than limitation the scope of the present invention, after the present invention has been read, various equivalences of the those skilled in the art to the present invention
The modification of form falls within the application appended claims limited range.
To effectively improve short-term wind-electricity power precision of prediction, the inventive method is carried out pre- using EWT to original wind power
Processing, is broken down into the empirical mode of series of features inequality, forecast model is established to each empirical mode.Experience small echo sheet
It is the one group of bandpass filter selected according to signal spectrum characteristic in matter, so as to adaptively filter out tune from primary signal
Width-frequency modulation composition.In order to determine the frequency range of bandpass filter, adaptivenon-uniform sampling is carried out to the Fourier spectrums of signal first.
Illustrate EWT adaptive decomposition processes with reference to Fig. 1.According to Shannon criterion, Fourier supports are defined as [0, π] and vacation
Surely N number of continuous part is divided into, makes Λn=[wn-1,wn] represent each segmentation border.Wherein:N=1,2, L, N, w0=
0,wN=π, wnThe midpoint being chosen between the two neighboring maximum point of signal Fourier spectrums, it is obviousWith every
Individual wnCentered on, definition width is Tn=2 τnTransitional region, as shown in dash area in figure.In segmentation section ΛnOn, definition
Experience small echo is each ΛnOn bandpass filter, and according to Meyer wavelet construction methods construct experience small echo.Gilles structures
The experience wavelet function made is
Experience scaling function is
In formula:τn=γ wn;When, it ensure thatFor tight frame.Generally, letter
Number β (x) is defined as β (x)=x4(35-84x+70x2-20x3)。
So as to which primary signal can be reconfigured as
In formula:* it is convolution algorithm;For Coefficients of Approximation;For x (t) experience wavelet transformation.
Empirical mode xk(t) define as the following formula
Standard regression analysis is under the conditions of given X=x, and dependent variable Y is obtained by minimizing square error loss function
Conditional mean E (Y | X=x) estimation, but this method can only provide the one-sided information of dependent variable Y conditions distribution, have ignored it
His information.In addition, when Y is when being distributed as containing singular value in thick tail or data, Regression Analysis Result robustness is poor.And divide
Digit recurrence is that dependent variable Y condition quantile returns to independent variable X, so as to obtain all quantile Regressions prediction moulds
Type.Therefore, quantile estimate more can accurately describe changes of the independent variable X to dependent variable Y relative to common least square regression
The influence of scope and condition distribution shape.
Under specified criteria X=x, conditional distribution function is Y≤y cumulative probability, i.e.,
F (y | X=x)=P (Y≤y | X=x) (5)
α quantiles Qα(x) it is that Y is more than or equal to Q under the conditions of given X=xα(x) cumulative probability is exactly α, i.e.,
Qα(x)=inf { y:F (y | X=x) >=α } (6)
In general linear conditions quantile estimate is expressed as
QY(α | X=x)=β0(α)+β1(α)x1+β2(α)x2+L+βk(α)xk≡X′β(α) (7)
In formula:QY(α | X=x) is dependent variable Y in independent variable X=[x1,x2,L,xk] under the α condition quantile, quantile α
∈ (0,1), β (α) is regression coefficient vector, and it changes with quantile α change.
Condition quantile solves parameter vector β (α) estimate by minimizing loss function, defines loss function and is
So as to which quantile estimate can be converted into following optimization problem
Give some quantile α, pass through parameter vector estimate corresponding to solution, you can describe independent variable now to because
The influence of variable.Then when α is in the continuous value in (0,1) feasible section, you can obtain Y condition distribution.
QRF is the innovatory algorithm of Breiman random forests, by combine quantile estimate characteristic, so as to provide because
The full terms distributed intelligence of variable.QRF possesses theoretical foundation, is proved to simultaneously as a kind of nonparametric machine learning method
With uniformity.
RF is counted as an adaptability nearest neighbour classification and regression process, to each X=x, can obtain original n sight
Examine one weight set w of valuei(x), i=1,2, L, n.RF essence is the weighted sum by the use of all dependent variable observations as because becoming
Measure Y conditional means E (Y | X=x) estimation.In addition, QRF decision trees are produced with standard RF algorithms, condition distribution is to pass through sight
The dependent variable Weighted estimation measured obtains, wherein the weight of each observation is equal to RF algorithm weights.
Thus, QRF defines E (1{Y≤y}| X=x) be estimated as observation 1{Y≤y}Weighted average, i.e.,
QRF algorithms concretely comprise the following steps:
1) k decision tree T (θ is generatedt), t=1,2, L, k.To each leaf node of every decision tree, the leaf node institute is investigated
There is observation;
2) X=x is given, travels through all decision trees.Calculate the weight w of every decision tree observationi(x,θt), i ∈ 1,2,
L,n}.By to decision tree weight wi(x,θt), t=1,2, L, k are averaged to obtain the weight of each observation i ∈ { 1,2, L, n }
wi(x);
3) to all y ∈ R, the weight drawn using step 2), the estimation of distribution function is calculated by formula (10).
To each node of every decision tree, RF, which is returned, only to be retained the average of observation and have ignored other information, and QRF
Retain all observations in node, and calculate condition distribution on this basis.
The present invention obtains probability density prediction result using Density Estimator method from condition distribution.Density Estimator is
The nonparametric computational methods of its density function are estimated by the stochastic variable from same unknown distribution function of one group of observation.
If X1,X2,L,XnUnitary continuous population sample is taken from, the Density Estimator of the population density function f (x) at the x of arbitrfary point is determined
Justice is
In formula:K (x) is kernel function, the gaussian kernel function form that uses of the present invention forH is band
Wide coefficient, span are 1.8~2.0.
The shortcomings that being difficult to characterize changed power uncertainty completely for general wind power point prediction method, the present invention is built
EWT-QRF short-term wind-electricity power probability density forecast models have been found, any time wind power fluctuation range and general can be obtained
Rate density exports.First, original wind power sequence is pre-processed using EWT signal processing technologies, by its adaptive point
Solve as the empirical mode of some frequency inequalities.Task amount is modeled to reduce, empirical mode similar in frequency size is merged into newly
Component.Secondly, the merging of input variable collection is chosen to new component and establishes QRF forecast models, obtain the wind under different quantiles
Electrical power prediction result, and different component prediction results are superimposed to the condition distribution for obtaining predicted value.Finally, estimated using cuclear density
The output wind power probability density prediction of meter method.
Using wind power plant actual measurement wind power data as research object, the inventive method estimated performance is verified.Datum
It is 30min according to sampling time interval, original wind power sequence is decomposed using EWT methods, selected part result such as Fig. 2
It is shown.It can be seen that original wind power is decomposed into 11 empirical modes by EWT, each mode frequency feature is more
Substantially.Fig. 3 is each mode spectrum section boundaries value, and empirical mode is divided into three classes by the present invention according to frequency size, is respectively
Low frequency, intermediate frequency and high frequency, then merge the empirical mode in class, are reconstructed into low frequency component, intermediate frequency component and height respectively
Frequency component.QRF forecast models are established to each component, so as to reduce modeling work amount, improve efficiency.Fig. 4 reconstructs for each component
As a result.
To illustrate that EWT is used for the validity of wind power prediction, EMD methods are used for point of power data by the present invention simultaneously
Solution, and contrast prediction result.Fig. 5 and Fig. 6 is respectively EMD decomposition results and reconstruct component.
The selection of input variable has to model prediction performance to be directly affected, and the present invention is quantified using Pearson correlation coefficients
The correlation between variable is evaluated, and the larger input variable set of correlation is chosen from preceding 10 moment at moment to be predicted.
Preset time sequence, Pearson correlation coefficients have weighed xtWith xt-τBetween dependency relation, can effectively determine input variable collection
Close.Wherein, τ is lag order.Pearson correlation coefficients r spans are [- 1,1], and 1 represents perfect positive correlation, and -1 has represented
Complete negatively correlated, 0 represents non-correlation between variable.When | r | during > 0.3, that is, think exist between variable compared with strong correlation relation.It is to be predicted
Coefficient correlation calculation formula is between moment wind power and preceding τ moment power
In formula:xt,i、xt-τ,iRespectively variable xtAnd xt-τI-th of sample value;Respectively variable xtAnd xt-τ's
Average value;N is sample size.
Input variable set is chosen to the component that EWT and EMD is decomposed, as a result respectively as shown in Table 1 and Table 2.
Table 1EWT component Input variable selection results
Table 2EMD component Input variable selection results
Missed using mean absolute percentage error (mean absolute percentage error, MAPE) and root mean square
Poor (root mean square error, RMSE) is used as certainty point prediction modelling effect evaluation index, calculation formula difference
For
In formula:N is future position number;yi、Respectively i-th of future position power actual value and predicted value.
Using section coverage rate (interval coverage percentage, ICP) and section mean breadth
(interval average width, IAW) is used as probability interval forecast model evaluation index, and calculation formula is respectively
In formula:ξ1-αFor under given 1- α level of confidence real power value fall the number in confidential interval;uiFor i-th
The individual future position confidential interval upper limit, liFor i-th of future position lower limit of confidence interval.
30min predictions, and verify the inventive method estimated performance in advance are carried out to actual measurement wind power.
1) directly original power data are predicted.Quantile estimate forest parameters are arranged to:Decision tree number is
500, node minimum dimension is 5, and every decision tree randomly selects m from input variable settry=2M/3 variable carries out weight
Study, M choose with reference to table 1.To obtain condition distribution, it is 0.01~0.99 that the present invention, which sets quantile scope, step-length 0.01,
99 prediction results can be obtained to each future position.Meanwhile the present invention establishes BP, SVMs (support
Vector machines, SVM) contrast model.BP learning rates are 0.001, learning objective 0.01, iteration 10000 times, model
Structural parameters through test of many times relatively after be arranged to 9-15-1, wherein first parameter is input layer variable number, hidden layer god
It is 15 through first number, output layer neuron number is 1.SVM model learnings parameter C and ε passes through grid data service optimum option, ginseng
Number scopes are [- 8,8], iteration step length 1.Table 3 is BP, SVM and QRF model prediction result, and QRF models take 0.5 quantile bar
Predicted value under part.As can be seen that during using Individual forecast model, predicted value lags behind actual value variation tendency, hysteresis be present,
Cause larger prediction error.
2) EMD and EWT methods are respectively adopted resolution process is carried out to original power data, then each component is built respectively
Vertical forecast model.Built-up pattern effectively increases precision of prediction as can be seen from Table 3.In addition, EWT method performances are better than
EMD methods, EWT-BP, EWT-SVM, EWT-QRF are carried respectively relative to EMD-BP, EMD-SVM, EMD-QRF model M APE indexs
High 24.97%, 13.58% and 22.07%, RMSE index be respectively increased 27.20%, 19.37% and 29.63%.Wherein:
EMD-BP models random component, details coefficients and trend component structural parameters are followed successively by 4-11-1,8-16-1 and 10-17-1,
EWT-BP models high fdrequency component, intermediate frequency component and low frequency component structural parameters are followed successively by 5-9-1,7-15-1 and 10-15-1;SVM
With QRF parameter settings with 1).Fig. 7 is power interval prediction result under 80% confidence level, and EWT-QRF interval widths are narrower, excellent
Gesture is obvious, so as to be more beneficial for science decision.
The different model wind power prediction result statistics of table 3
The condition distribution obtained to QRF, EMD-QRF and EWT-QRF can provide any time wind using Density Estimator
Electrical power probability density prediction result, choose one day at different moments prediction result as shown in Fig. 8-Figure 13.As can be seen that EWT-
For QRF models with high probability close to actual value, probability density curve is taller and thinner, and fluctuation range is more concentrated, and is advantageous in narrower model
Educated decisions are made in enclosing.
In summary, the short-term wind-electricity power probability density Forecasting Methodology of a kind of EWT quantile estimates forest of the invention
With following advantage:1) using EWT, this NEW ADAPTIVE signal processing method is carried out at decomposition to original wind power sequence
Reason, it is broken down into the empirical mode of multiple frequecy characteristic inequalities.Relative to EMD methods, the component that EWT is decomposed is with more explanation
Meaning.Secondly, the EWT-QRF forecast models of foundation have more preferable precision of prediction relative to EMD-QRF models, and MAPE indexs carry
High 29.31%, RMSE indexs improve 29.63%.Under same confidence level, interval width is narrower, is advantageous to make determining for science
Plan.2) quantile estimate forest model can provide the prediction result under any quantile, so as to obtain the distribution of predicted value condition,
And further obtain the prediction of wind power probability density.Quantile estimate forest as a kind of Nonparametric Estimation, have by
Model parameter influences the advantages of small, strong robustness, less amount of calculation, is suitably applied the prediction of short-term wind-electricity power probability density.
The inventive method arranges power system wind-powered electricity generation, and operation plan and guarantee power network safety operation have one a few days ago
Fixed reference value.
Claims (7)
- A kind of 1. short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimates forest, it is characterised in that:Including following Step:(1) master data needed for power system short-term wind power prediction is obtained, and initial data is pre-processed, is carried out Abnormality value removing or amendment;(2) pretreated wind power sequence is decomposed into the empirical mode of series of features inequality using EWT;(3) empirical mode is reconfigured according to frequency range, forms high fdrequency component, intermediate frequency component and low frequency component;(4) input variable set is chosen using Pearson correlation coefficients to each component;(5) quantile estimate forest model is established to each component, obtains different quantile regression forecasting results;(6) each component prediction result is superimposed, obtains final wind power prediction value;(7) prediction of wind power probability density is obtained using Density Estimator, and by surveying wind power data verification this hair The validity of bright method.
- 2. the short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimates forest according to claim 1, it is special Sign is:Original wind power sequence is decomposed into a series of Empirical Mode of frequecy characteristic inequalities using EWT technologies by step (2) Formula, the specific calculating process of EWT methods are:2.1 determine the frequency range of bandpass filter;Adaptivenon-uniform sampling, definition are carried out to the Fourier spectrums of signal first Fourier supports is [0, π] and assume to be divided into N number of continuous part, make Λn=[wn-1,wn] represent each segmentation side Boundary;Wherein:N=1,2, L, N, w0=0, wN=π, wnIn being chosen between the two neighboring maximum point of signal Fourier spectrums Point, it is obvious2.2 with each wnCentered on, definition width is Tn=2 τnTransitional region;2.3 in segmentation section ΛnOn, it is each Λ to define experience small echonOn bandpass filter, and according to Meyer Construction of Wavelets Method construct experience small echo, the experience wavelet function generally constructed are<mrow> <msub> <mover> <mi>&psi;</mi> <mo>^</mo> </mover> <mi>n</mi> </msub> <mrow> <mo>(</mo> <mi>w</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <mn>1</mn> </mtd> <mtd> <mrow> <msub> <mi>w</mi> <mi>n</mi> </msub> <mo>+</mo> <msub> <mi>&tau;</mi> <mi>n</mi> </msub> <mo>&le;</mo> <mrow> <mo>|</mo> <mi>w</mi> <mo>|</mo> </mrow> <mo>&le;</mo> <msub> <mi>w</mi> <mrow> <mi>n</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>-</mo> <msub> <mi>&tau;</mi> <mrow> <mi>n</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mi>cos</mi> <mo>&lsqb;</mo> <mfrac> <mi>&pi;</mi> <mn>2</mn> </mfrac> <mi>&beta;</mi> <mrow> <mo>(</mo> <mfrac> <mn>1</mn> <mrow> <mn>2</mn> <msub> <mi>&tau;</mi> <mrow> <mi>n</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> </mrow> </mfrac> <mo>(</mo> <mrow> <mrow> <mo>|</mo> <mi>w</mi> <mo>|</mo> </mrow> <mo>-</mo> <msub> <mi>w</mi> <mrow> <mi>n</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>+</mo> <msub> <mi>&tau;</mi> <mrow> <mi>n</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> </mrow> <mo>)</mo> <mo>)</mo> </mrow> <mo>&rsqb;</mo> </mrow> </mtd> <mtd> <mrow> <msub> <mi>w</mi> <mrow> <mi>n</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>-</mo> <msub> <mi>&tau;</mi> <mrow> <mi>n</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>&le;</mo> <mrow> <mo>|</mo> <mi>w</mi> <mo>|</mo> </mrow> <mo>&le;</mo> <msub> <mi>w</mi> <mrow> <mi>n</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>+</mo> <msub> <mi>&tau;</mi> <mrow> <mi>n</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mi>sin</mi> <mo>&lsqb;</mo> <mfrac> <mi>&pi;</mi> <mn>2</mn> </mfrac> <mi>&beta;</mi> <mrow> <mo>(</mo> <mfrac> <mn>1</mn> <mrow> <mn>2</mn> <msub> <mi>&tau;</mi> <mi>n</mi> </msub> </mrow> </mfrac> <mo>(</mo> <mrow> <mrow> <mo>|</mo> <mi>w</mi> <mo>|</mo> </mrow> <mo>-</mo> <msub> <mi>w</mi> <mi>n</mi> </msub> <mo>+</mo> <msub> <mi>&tau;</mi> <mi>n</mi> </msub> </mrow> <mo>)</mo> <mo>)</mo> </mrow> <mo>&rsqb;</mo> </mrow> </mtd> <mtd> <mrow> <msub> <mi>w</mi> <mi>n</mi> </msub> <mo>-</mo> <msub> <mi>&tau;</mi> <mi>n</mi> </msub> <mo>&le;</mo> <mrow> <mo>|</mo> <mi>w</mi> <mo>|</mo> </mrow> <mo>&le;</mo> <msub> <mi>w</mi> <mi>n</mi> </msub> <mo>+</mo> <msub> <mi>&tau;</mi> <mi>n</mi> </msub> </mrow> </mtd> </mtr> <mtr> <mtd> <mn>0</mn> </mtd> <mtd> <mrow> <mi>o</mi> <mi>t</mi> <mi>h</mi> <mi>e</mi> <mi>r</mi> <mi>w</mi> <mi>i</mi> <mi>s</mi> <mi>e</mi> </mrow> </mtd> </mtr> </mtable> </mfenced> </mrow>Experience scaling function is<mrow> <msub> <mover> <mi>&phi;</mi> <mo>^</mo> </mover> <mi>n</mi> </msub> <mrow> <mo>(</mo> <mi>w</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <mn>1</mn> </mtd> <mtd> <mrow> <mrow> <mo>|</mo> <mi>w</mi> <mo>|</mo> </mrow> <mo>&le;</mo> <msub> <mi>w</mi> <mi>n</mi> </msub> <mo>-</mo> <msub> <mi>&tau;</mi> <mi>n</mi> </msub> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mi>c</mi> <mi>o</mi> <mi>s</mi> <mo>&lsqb;</mo> <mfrac> <mi>&pi;</mi> <mn>2</mn> </mfrac> <mi>&beta;</mi> <mrow> <mo>(</mo> <mfrac> <mn>1</mn> <mrow> <mn>2</mn> <msub> <mi>&tau;</mi> <mi>n</mi> </msub> </mrow> </mfrac> <mo>(</mo> <mrow> <mrow> <mo>|</mo> <mi>w</mi> <mo>|</mo> </mrow> <mo>-</mo> <msub> <mi>w</mi> <mi>n</mi> </msub> <mo>+</mo> <msub> <mi>&tau;</mi> <mi>n</mi> </msub> </mrow> <mo>)</mo> <mo>)</mo> </mrow> <mo>&rsqb;</mo> </mrow> </mtd> <mtd> <mrow> <msub> <mi>w</mi> <mi>n</mi> </msub> <mo>-</mo> <msub> <mi>&tau;</mi> <mi>n</mi> </msub> <mo>&le;</mo> <mrow> <mo>|</mo> <mi>w</mi> <mo>|</mo> </mrow> <mo>&le;</mo> <msub> <mi>w</mi> <mi>n</mi> </msub> <mo>+</mo> <msub> <mi>&tau;</mi> <mi>n</mi> </msub> </mrow> </mtd> </mtr> <mtr> <mtd> <mn>0</mn> </mtd> <mtd> <mrow> <mi>o</mi> <mi>t</mi> <mi>h</mi> <mi>e</mi> <mi>r</mi> <mi>w</mi> <mi>i</mi> <mi>s</mi> <mi>e</mi> </mrow> </mtd> </mtr> </mtable> </mfenced> </mrow>In formula:τn=γ wn;When, it ensure thatFor tight frame;Generally, function β (x) it is defined as β (x)=x4(35-84x+70x2-20x3);2.4 so as to which primary signal can be reconfigured as<mrow> <mi>x</mi> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>=</mo> <msubsup> <mi>W</mi> <mi>x</mi> <mi>&epsiv;</mi> </msubsup> <mrow> <mo>(</mo> <mn>0</mn> <mo>,</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>*</mo> <msub> <mi>&phi;</mi> <mn>1</mn> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>+</mo> <munderover> <mo>&Sigma;</mo> <mrow> <mi>n</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <msubsup> <mi>W</mi> <mi>x</mi> <mi>&epsiv;</mi> </msubsup> <mrow> <mo>(</mo> <mi>n</mi> <mo>,</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>*</mo> <msub> <mi>&phi;</mi> <mi>n</mi> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> </mrow>In formula:* it is convolution algorithm;For Coefficients of Approximation;For x (t) experience wavelet transformation;Empirical mode xk(t) it is defined as<mrow> <mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>(</mo> <mi>t</mi> <mo>)</mo> <mo>=</mo> <msubsup> <mi>W</mi> <mi>x</mi> <mi>&epsiv;</mi> </msubsup> <mo>(</mo> <mn>0</mn> <mo>,</mo> <mi>t</mi> <mo>)</mo> <mo>*</mo> <msub> <mi>&phi;</mi> <mn>1</mn> </msub> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mtd> </mtr> <mtr> <mtd> <msub> <mi>x</mi> <mi>k</mi> </msub> <mo>(</mo> <mi>t</mi> <mo>)</mo> <mo>=</mo> <msubsup> <mi>W</mi> <mi>x</mi> <mi>&epsiv;</mi> </msubsup> <mo>(</mo> <mi>n</mi> <mo>,</mo> <mi>t</mi> <mo>)</mo> <mo>*</mo> <msub> <mi>&phi;</mi> <mi>n</mi> </msub> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mtd> </mtr> </mtable> </mfenced> <mo>.</mo> </mrow>
- 3. the short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimates forest according to claim 1, it is special Sign is:Step (3) is reconfigured empirical mode according to frequency range, forms high fdrequency component, intermediate frequency component and low frequency Component, the frequency range demarcation method are specially:Empirical mode in [0, π] frequency range is divided into three parts:Frequency It is low frequency component less than 1, frequency is more than 1 and is more than 2 as high fdrequency component less than 2 for intermediate frequency component, frequency.
- 4. the short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimates forest according to claim 1, it is special Sign is:Step (4) chooses input variable set, the Pearson correlation coefficients tool to each component using Pearson correlation coefficients Body computational methods are:Preset time sequence, Pearson correlation coefficients have weighed xtWith xt-τBetween dependency relation, can effectively determine input become Duration set;Wherein, τ is lag order;Pearson correlation coefficients r spans are [- 1,1], and 1 represents perfect positive correlation, -1 table Show perfect negative correlation, 0 represents non-correlation between variable;When | r | during > 0.3, that is, think exist between variable compared with strong correlation relation;Treat Coefficient correlation calculation formula is between prediction time wind power and preceding τ moment power<mrow> <mi>r</mi> <mo>=</mo> <mfrac> <mrow> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mrow> <mi>t</mi> <mo>,</mo> <mi>i</mi> </mrow> </msub> <mo>-</mo> <msub> <mover> <mi>x</mi> <mo>&OverBar;</mo> </mover> <mi>t</mi> </msub> <mo>)</mo> </mrow> <mo>&CenterDot;</mo> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mrow> <mi>t</mi> <mo>-</mo> <mi>&tau;</mi> <mo>,</mo> <mi>i</mi> </mrow> </msub> <mo>-</mo> <msub> <mover> <mi>x</mi> <mo>&OverBar;</mo> </mover> <mrow> <mi>t</mi> <mo>-</mo> <mi>&tau;</mi> </mrow> </msub> <mo>)</mo> </mrow> </mrow> <mrow> <msqrt> <mrow> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <msup> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mrow> <mi>t</mi> <mo>,</mo> <mi>i</mi> </mrow> </msub> <mo>-</mo> <msub> <mover> <mi>x</mi> <mo>&OverBar;</mo> </mover> <mi>t</mi> </msub> <mo>)</mo> </mrow> <mn>2</mn> </msup> </mrow> </msqrt> <mo>&CenterDot;</mo> <msqrt> <mrow> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <msup> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mrow> <mi>t</mi> <mo>-</mo> <mi>&tau;</mi> <mo>,</mo> <mi>i</mi> </mrow> </msub> <mo>-</mo> <msub> <mover> <mi>x</mi> <mo>&OverBar;</mo> </mover> <mrow> <mi>t</mi> <mo>-</mo> <mi>&tau;</mi> </mrow> </msub> <mo>)</mo> </mrow> <mn>2</mn> </msup> </mrow> </msqrt> </mrow> </mfrac> </mrow>In formula:xt,i、xt-τ,iRespectively variable xtAnd xt-τI-th of sample value;Respectively variable xtAnd xt-τBe averaged Value;N is sample size.
- 5. the short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimates forest according to claim 1, it is special Sign is:Step (5) establishes quantile estimate forest model to each component, obtains different quantile regression forecasting results;Divide position Number returns the advantage that forest model combines quantile estimate and Random Forest model, can provide different quantile regression forecastings As a result;As a kind of nonparametric ensemble machine learning method, QRF have concurrently arithmetic speed is fast, model performance by parameter influenceed it is small, compared with Strong the advantages that holding making an uproar property;The specific calculating process of quantile estimate is:5.1 under specified criteria X=x, and conditional distribution function is Y≤y cumulative probability, i.e.,F (y | X=x)=P (Y≤y | X=x)α quantiles Qα(x) it is that Y is more than or equal to Q under the conditions of given X=xα(x) cumulative probability is exactly α, i.e.,Qα(x)=inf { y:F (y | X=x) >=α }5.2 in general linear conditions quantile estimates are expressed asQY(α | X=x)=β0(α)+β1(α)x1+β2(α)x2+L+βk(α)xk≡X′β(α)In formula:QY(α | X=x) is dependent variable Y in independent variable X=[x1,x2,L,xk] under the α condition quantile, quantile α ∈ (0,1), β (α) is regression coefficient vector, and it changes with quantile α change;5.3 condition quantiles solve parameter vector β (α) estimate by minimizing loss function, define loss function and are<mrow> <msub> <mi>L</mi> <mi>&alpha;</mi> </msub> <mrow> <mo>(</mo> <mi>y</mi> <mo>,</mo> <msup> <mi>X</mi> <mo>&prime;</mo> </msup> <mi>&beta;</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <mi>&alpha;</mi> <mrow> <mo>|</mo> <mrow> <mi>y</mi> <mo>-</mo> <msup> <mi>X</mi> <mo>&prime;</mo> </msup> <mi>&beta;</mi> </mrow> <mo>|</mo> </mrow> <mo>,</mo> <mi>y</mi> <mo>></mo> <msup> <mi>X</mi> <mo>&prime;</mo> </msup> <mi>&beta;</mi> </mtd> </mtr> <mtr> <mtd> <mo>(</mo> <mn>1</mn> <mo>-</mo> <mi>&alpha;</mi> <mo>)</mo> <mrow> <mo>|</mo> <mrow> <mi>y</mi> <mo>-</mo> <msup> <mi>X</mi> <mo>&prime;</mo> </msup> <mi>&beta;</mi> </mrow> <mo>|</mo> </mrow> <mo>,</mo> <mi>y</mi> <mo>&le;</mo> <msup> <mi>X</mi> <mo>&prime;</mo> </msup> <mi>&beta;</mi> </mtd> </mtr> </mtable> </mfenced> </mrow>5.4 can be converted into following optimization problem so as to quantile estimate<mrow> <munder> <mrow> <mi>m</mi> <mi>i</mi> <mi>n</mi> </mrow> <mi>&beta;</mi> </munder> <munder> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>|</mo> <msub> <mi>Y</mi> <mi>i</mi> </msub> <mo>&GreaterEqual;</mo> <msubsup> <mi>X</mi> <mi>i</mi> <mo>&prime;</mo> </msubsup> <mi>&beta;</mi> </mrow> </munder> <mi>&alpha;</mi> <mo>|</mo> <mrow> <msub> <mi>Y</mi> <mi>i</mi> </msub> <mo>-</mo> <msubsup> <mi>X</mi> <mi>i</mi> <mo>&prime;</mo> </msubsup> <mi>&beta;</mi> </mrow> <mo>|</mo> <mo>+</mo> <munder> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>|</mo> <msub> <mi>Y</mi> <mi>i</mi> </msub> <mo><</mo> <msubsup> <mi>X</mi> <mi>i</mi> <mo>&prime;</mo> </msubsup> <mi>&beta;</mi> </mrow> </munder> <mrow> <mo>(</mo> <mn>1</mn> <mo>-</mo> <mi>&alpha;</mi> <mo>)</mo> </mrow> <mo>|</mo> <mrow> <msub> <mi>Y</mi> <mi>i</mi> </msub> <mo>-</mo> <msubsup> <mi>X</mi> <mi>i</mi> <mo>&prime;</mo> </msubsup> <mi>&beta;</mi> </mrow> <mo>|</mo> </mrow>Some quantile α is given, passes through parameter vector estimate corresponding to solution, you can describe independent variable now to dependent variable Influence;Then when α is in the continuous value in (0,1) feasible section, you can obtain Y condition distribution.
- 6. the short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimates forest according to claim 1, it is special Sign is:Step (5) establishes quantile estimate forest model to each component, obtains different quantile regression forecasting results;It is described The specific calculating process of quantile estimate forest is:6.1 random forests are counted as an adaptability nearest neighbour classification and regression process, to each X=x, can obtain original One weight set w of n observed valuei(x), i=1,2, L, n;Random forest essence is added using all dependent variable observations Power and the estimation as dependent variable Y conditional means E (Y | X=x);In addition, QRF decision trees are produced with standard RF algorithms, condition Distribution is obtained by the dependent variable Weighted estimation observed, wherein the weight of each observation is equal to RF algorithm weights;6.2 thus, and QRF defines E (1{Y≤y}| X=x) be estimated as observation 1{Y≤y}Weighted average, i.e.,<mrow> <mover> <mi>F</mi> <mo>^</mo> </mover> <mrow> <mo>(</mo> <mi>y</mi> <mo>|</mo> <mi>X</mi> <mo>=</mo> <mi>x</mi> <mo>)</mo> </mrow> <mo>=</mo> <munderover> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <msub> <mi>w</mi> <mi>i</mi> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> <msub> <mn>1</mn> <mrow> <mo>{</mo> <mi>Y</mi> <mo>&le;</mo> <mi>y</mi> <mo>}</mo> </mrow> </msub> </mrow>6.3 k decision tree T (θ of generationt), t=1,2, L, k;To each leaf node of every decision tree, all sights of the leaf node are investigated Measured value;6.4 given X=x, travel through all decision trees;Calculate the weight w of every decision tree observationi(x,θt), i ∈ 1,2, L, n};By to decision tree weight wi(x,θt), t=1,2, L, k are averaged to obtain the weight w of each observation i ∈ { 1,2, L, n }i (x);6.5 couples of all y ∈ R, the weight drawn using step 6.4, you can calculate the estimation of distribution function;To every decision tree Each node, RF, which is returned, only to be retained the average of observation and have ignored other information, and QRF retains all observations in node, And condition distribution is calculated on this basis.
- 7. the short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimates forest according to claim 1, it is special Sign is:The prediction of wind power probability density is obtained using Density Estimator in step (7), and by surveying wind power data Verify the validity of the inventive method;The specific calculating process of Density Estimator is:Density Estimator is to estimate its density function by the stochastic variable from same unknown distribution function of one group of observation Nonparametric computational methods;If X1,X2,L,XnUnitary continuous population sample is taken from, the population density function f at the x of arbitrfary point (x) Density Estimator is defined as<mrow> <msub> <mover> <mi>f</mi> <mo>^</mo> </mover> <mi>h</mi> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mrow> <mi>n</mi> <mi>h</mi> </mrow> </mfrac> <munderover> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <mi>K</mi> <mrow> <mo>(</mo> <mfrac> <mrow> <mi>x</mi> <mo>-</mo> <msub> <mi>X</mi> <mi>i</mi> </msub> </mrow> <mi>h</mi> </mfrac> <mo>)</mo> </mrow> </mrow>In formula:K (x) is kernel function, the gaussian kernel function form used herein forH is bandwidth system Number, span are 1.8~2.0.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710850654.7A CN107704953A (en) | 2017-09-20 | 2017-09-20 | The short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimate forests |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710850654.7A CN107704953A (en) | 2017-09-20 | 2017-09-20 | The short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimate forests |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107704953A true CN107704953A (en) | 2018-02-16 |
Family
ID=61171643
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710850654.7A Pending CN107704953A (en) | 2017-09-20 | 2017-09-20 | The short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimate forests |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107704953A (en) |
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108416695A (en) * | 2018-02-24 | 2018-08-17 | 合肥工业大学 | Electric load probability density prediction technique based on deep learning and system, medium |
CN108460501A (en) * | 2018-05-10 | 2018-08-28 | 湖北工业大学 | A kind of wind power station output power predicting method based on built-up pattern |
CN108846517A (en) * | 2018-06-12 | 2018-11-20 | 清华大学 | A kind of probability short-term electric load prediction integrated approach of quantile |
CN108984797A (en) * | 2018-08-20 | 2018-12-11 | 新疆工程学院 | A kind of method for quick estimating of magnanimity monitor-type long-term sequence data related coefficient |
CN109085481A (en) * | 2018-10-23 | 2018-12-25 | 国网江苏省电力有限公司徐州供电分公司 | A kind of cable local discharge localization method based on EWT |
CN109086793A (en) * | 2018-06-27 | 2018-12-25 | 东北大学 | A kind of abnormality recognition method of wind-driven generator |
CN109101390A (en) * | 2018-06-29 | 2018-12-28 | 平安科技(深圳)有限公司 | Timed task abnormality monitoring method, electronic device and medium based on Gaussian Profile |
CN109117992A (en) * | 2018-07-27 | 2019-01-01 | 华北电力大学 | Ultra-short term wind power prediction method based on WD-LA-WRF model |
CN109597967A (en) * | 2018-11-20 | 2019-04-09 | 江苏云上电力科技有限公司 | A kind of electric system distribution transforming power station load data abnormality detection and restorative procedure |
CN110796303A (en) * | 2019-10-24 | 2020-02-14 | 国网安徽省电力有限公司电力科学研究院 | Short-term power load prediction method based on EWT and ODBSCAN |
CN111612262A (en) * | 2020-01-15 | 2020-09-01 | 长沙理工大学 | Wind power probability prediction method based on quantile regression |
CN112183848A (en) * | 2020-09-25 | 2021-01-05 | 合肥工业大学 | Power load probability prediction method based on DWT-SVQR integration |
CN112561414A (en) * | 2019-09-25 | 2021-03-26 | 顺丰科技有限公司 | Inventory management method, device and computer readable storage medium |
CN112926801A (en) * | 2021-03-31 | 2021-06-08 | 云南电网有限责任公司 | Load curve combined prediction method and device based on quantile regression |
CN113205223A (en) * | 2021-05-17 | 2021-08-03 | 上海电力大学 | Electric quantity prediction system and prediction method thereof |
CN113258565A (en) * | 2021-05-11 | 2021-08-13 | 广东电网有限责任公司韶关供电局 | Frequency adjusting method, device, equipment and storage medium |
CN113469467A (en) * | 2021-09-02 | 2021-10-01 | 国能日新科技股份有限公司 | Wind power ultra-short term prediction method and device based on band-pass filtering |
CN113536584A (en) * | 2021-07-23 | 2021-10-22 | 河北工业大学 | Wind speed reverse generation method based on probability model and multi-time window self-correlation verification |
CN114548464A (en) * | 2020-11-19 | 2022-05-27 | 国网湖南省电力有限公司 | Wind power confidence interval prediction method, device and medium based on wavelet transformation |
CN115409296A (en) * | 2022-11-02 | 2022-11-29 | 国网江西省电力有限公司电力科学研究院 | Method for predicting net load probability of active power distribution network |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106067073A (en) * | 2016-05-30 | 2016-11-02 | 都城绿色能源有限公司 | A kind of wind power forecasting method based on wind-resources correlation analysis |
CN106295899A (en) * | 2016-08-17 | 2017-01-04 | 合肥工业大学 | Based on genetic algorithm and the wind power probability density Forecasting Methodology supporting vector quantile estimate |
CN106548253A (en) * | 2016-11-08 | 2017-03-29 | 中国地质大学(武汉) | Method based on the wind power prediction of nonparametric probability |
CN106846173A (en) * | 2016-12-30 | 2017-06-13 | 国网新疆电力公司电力科学研究院 | Short-term wind power forecast method based on EWT ESN |
-
2017
- 2017-09-20 CN CN201710850654.7A patent/CN107704953A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106067073A (en) * | 2016-05-30 | 2016-11-02 | 都城绿色能源有限公司 | A kind of wind power forecasting method based on wind-resources correlation analysis |
CN106295899A (en) * | 2016-08-17 | 2017-01-04 | 合肥工业大学 | Based on genetic algorithm and the wind power probability density Forecasting Methodology supporting vector quantile estimate |
CN106548253A (en) * | 2016-11-08 | 2017-03-29 | 中国地质大学(武汉) | Method based on the wind power prediction of nonparametric probability |
CN106846173A (en) * | 2016-12-30 | 2017-06-13 | 国网新疆电力公司电力科学研究院 | Short-term wind power forecast method based on EWT ESN |
Non-Patent Citations (6)
Title |
---|
BENJAMIN VOLZ 等: "Predicting pedestrian crossing using Quantile Regression forests", 《2016 IEEE INTELLIGENT VEHICLES SYMPOSIUM (IV)》 * |
HAIXIANG ZANG 等: "Short-term wind speed forecasting based on an EEMD-CAPSO-RVM model", 《2016 IEEE PES ASIA-PACIFIC POWER AND ENERGY ENGINEERING CONFERENCE (APPEEC)》 * |
JIANMING HU 等: "Short-term wind speed prediction using empirical wavelet transform and Gaussian process regression", 《ENERGY》 * |
NICOLAI MEINSHAUSEN 等: "Quantile Regression Forests", 《JOURNAL OF MACHINE LEARNING RESEARCH》 * |
YAO ZHANG 等: "Review on probabilistic forecasting of wind power generation", 《RENEWABLE AND SUSTAINABLE ENERGY REVIEWS》 * |
何耀耀 等: "基于RBF神经网络分位数回归的电力负荷概率密度预测方法", 《中国电机工程学报》 * |
Cited By (32)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108416695B (en) * | 2018-02-24 | 2020-07-07 | 合肥工业大学 | Power load probability density prediction method, system and medium based on deep learning |
CN108416695A (en) * | 2018-02-24 | 2018-08-17 | 合肥工业大学 | Electric load probability density prediction technique based on deep learning and system, medium |
CN108460501A (en) * | 2018-05-10 | 2018-08-28 | 湖北工业大学 | A kind of wind power station output power predicting method based on built-up pattern |
CN108460501B (en) * | 2018-05-10 | 2021-09-14 | 湖北工业大学 | Wind power station output power prediction method based on combined model |
CN108846517A (en) * | 2018-06-12 | 2018-11-20 | 清华大学 | A kind of probability short-term electric load prediction integrated approach of quantile |
CN108846517B (en) * | 2018-06-12 | 2021-03-16 | 清华大学 | Integration method for predicating quantile probabilistic short-term power load |
CN109086793B (en) * | 2018-06-27 | 2021-11-16 | 东北大学 | Abnormity identification method for wind driven generator |
CN109086793A (en) * | 2018-06-27 | 2018-12-25 | 东北大学 | A kind of abnormality recognition method of wind-driven generator |
CN109101390B (en) * | 2018-06-29 | 2021-08-24 | 平安科技(深圳)有限公司 | Timed task abnormity monitoring method based on Gaussian distribution, electronic device and medium |
CN109101390A (en) * | 2018-06-29 | 2018-12-28 | 平安科技(深圳)有限公司 | Timed task abnormality monitoring method, electronic device and medium based on Gaussian Profile |
CN109117992B (en) * | 2018-07-27 | 2022-01-04 | 华北电力大学 | Ultra-short-term wind power prediction method based on WD-LA-WRF model |
CN109117992A (en) * | 2018-07-27 | 2019-01-01 | 华北电力大学 | Ultra-short term wind power prediction method based on WD-LA-WRF model |
CN108984797A (en) * | 2018-08-20 | 2018-12-11 | 新疆工程学院 | A kind of method for quick estimating of magnanimity monitor-type long-term sequence data related coefficient |
CN109085481B (en) * | 2018-10-23 | 2021-11-26 | 国网江苏省电力有限公司徐州供电分公司 | Cable partial discharge positioning method based on EWT |
CN109085481A (en) * | 2018-10-23 | 2018-12-25 | 国网江苏省电力有限公司徐州供电分公司 | A kind of cable local discharge localization method based on EWT |
CN109597967B (en) * | 2018-11-20 | 2023-08-25 | 江苏云上电力科技有限公司 | Power system distribution substation load data anomaly detection and repair method |
CN109597967A (en) * | 2018-11-20 | 2019-04-09 | 江苏云上电力科技有限公司 | A kind of electric system distribution transforming power station load data abnormality detection and restorative procedure |
CN112561414A (en) * | 2019-09-25 | 2021-03-26 | 顺丰科技有限公司 | Inventory management method, device and computer readable storage medium |
CN110796303B (en) * | 2019-10-24 | 2023-11-07 | 国网安徽省电力有限公司电力科学研究院 | Short-term power load prediction method based on EWT and ODBSCAN |
CN110796303A (en) * | 2019-10-24 | 2020-02-14 | 国网安徽省电力有限公司电力科学研究院 | Short-term power load prediction method based on EWT and ODBSCAN |
CN111612262A (en) * | 2020-01-15 | 2020-09-01 | 长沙理工大学 | Wind power probability prediction method based on quantile regression |
CN112183848B (en) * | 2020-09-25 | 2022-08-30 | 合肥工业大学 | Power load probability prediction method based on DWT-SVQR integration |
CN112183848A (en) * | 2020-09-25 | 2021-01-05 | 合肥工业大学 | Power load probability prediction method based on DWT-SVQR integration |
CN114548464A (en) * | 2020-11-19 | 2022-05-27 | 国网湖南省电力有限公司 | Wind power confidence interval prediction method, device and medium based on wavelet transformation |
CN112926801A (en) * | 2021-03-31 | 2021-06-08 | 云南电网有限责任公司 | Load curve combined prediction method and device based on quantile regression |
CN112926801B (en) * | 2021-03-31 | 2022-11-01 | 云南电网有限责任公司 | Load curve combined prediction method and device based on quantile regression |
CN113258565A (en) * | 2021-05-11 | 2021-08-13 | 广东电网有限责任公司韶关供电局 | Frequency adjusting method, device, equipment and storage medium |
CN113205223A (en) * | 2021-05-17 | 2021-08-03 | 上海电力大学 | Electric quantity prediction system and prediction method thereof |
CN113536584A (en) * | 2021-07-23 | 2021-10-22 | 河北工业大学 | Wind speed reverse generation method based on probability model and multi-time window self-correlation verification |
CN113469467B (en) * | 2021-09-02 | 2021-11-30 | 国能日新科技股份有限公司 | Wind power ultra-short term prediction method and device based on band-pass filtering |
CN113469467A (en) * | 2021-09-02 | 2021-10-01 | 国能日新科技股份有限公司 | Wind power ultra-short term prediction method and device based on band-pass filtering |
CN115409296A (en) * | 2022-11-02 | 2022-11-29 | 国网江西省电力有限公司电力科学研究院 | Method for predicting net load probability of active power distribution network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107704953A (en) | The short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimate forests | |
Dong et al. | Hourly energy consumption prediction of an office building based on ensemble learning and energy consumption pattern classification | |
Tascikaraoglu et al. | A review of combined approaches for prediction of short-term wind speed and power | |
De Giorgi et al. | Assessment of the benefits of numerical weather predictions in wind power forecasting based on statistical methods | |
Wang et al. | A combined framework based on data preprocessing, neural networks and multi-tracker optimizer for wind speed prediction | |
CN103729550B (en) | Multiple-model integration Flood Forecasting Method based on propagation time cluster analysis | |
Wang et al. | Wind speed deterministic forecasting and probabilistic interval forecasting approach based on deep learning, modified tunicate swarm algorithm, and quantile regression | |
CN108038580A (en) | The multi-model integrated Forecasting Methodology of photovoltaic power based on synchronous extruding wavelet transformation | |
CN106875033A (en) | A kind of wind-powered electricity generation cluster power forecasting method based on dynamic self-adapting | |
CN105868853B (en) | Method for predicting short-term wind power combination probability | |
CN106295899B (en) | Wind power probability density Forecasting Methodology based on genetic algorithm Yu supporting vector quantile estimate | |
CN106846173A (en) | Short-term wind power forecast method based on EWT ESN | |
CN104573879A (en) | Photovoltaic power station output predicting method based on optimal similar day set | |
CN107341581A (en) | A kind of new energy output short term prediction method returned based on experience wavelet transformation and Gaussian process | |
CN110309603A (en) | A kind of short-term wind speed forecasting method and system based on wind speed characteristics | |
Zhou et al. | Short-term wind power prediction optimized by multi-objective dragonfly algorithm based on variational mode decomposition | |
CN103258142A (en) | Wind power plant short-term wind speed forecasting method based on weather patterns | |
CN103559540A (en) | Air speed ultra-short period on-line predication method based on adaptive neural-fuzzy inference system | |
Ismail et al. | A hybrid model of self organizing maps and least square support vector machine for river flow forecasting | |
CN107844849A (en) | A kind of new energy output short term prediction method returned based on experience wavelet transformation with improving Gaussian process | |
CN105005708B (en) | A kind of broad sense load Specialty aggregation method based on AP clustering algorithms | |
CN103235984B (en) | Longitudinal moment probability distribution computing method of output of wind electric field | |
CN105069236A (en) | Generalized load joint probability modeling method considering node spatial correlation of wind power plant | |
CN107203827A (en) | A kind of wind turbine forecasting wind speed optimization method based on multiscale analysis | |
Bing et al. | A hybrid short-term traffic flow multistep prediction method based on variational mode decomposition and long short-term memory model |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180216 |